Upload
nguyentruc
View
215
Download
2
Embed Size (px)
Citation preview
The role of the tongue and
pharynx in enhancement of vowel nasalization:
A real-time MRI investigation
of French nasal vowels.
Christopher Carignana,b, Ryan Shostedb, Maojing Fub, Zhi-Pei Liangb, Bradley Suttonb
aNorth Carolina State UniversitybUniversity of Illinois at Urbana-Champaign
INTERSPEECH 2013, August 29
Introduction: nasalization
� Nasal vowels are characterized by some degree of coupling between the nasal and oral cavities.� Velopharyngeal coupling, i.e., “nasalization”.
� It is well known that nasalization significantly alters the acoustic spectrum of vowels (Hawkins & Stevens, 1985; Katakoa et al., 2001; Pruthi et al., 2007).
� Aside from acoustic effects such as formant amplitude reduction and bandwidth widening, modeling and acoustic studies suggest that the F1/F2 frequencies modulated.
INTERSPEECH 2013, August 29
Introduction: effect on F1/F2
� The frequencies are centralized along the F1
dimension (i.e, F1 is raised for high vowels and lowered for low vowels). (Fujimura & Lindqvist, 1971; Diehl et al., 1991; Serrurier & Badin, 2008; Feng & Castelli, 1996)
� The F2 frequency is lowered for all non-back vowels (Serrurier & Badin, 2008; Feng & Castelli, 1996; Carignan, 2013).
INTERSPEECH 2013, August 29
Introduction: F1/F2 (cont.)
� Krakow et al. (1988) observed that the F1 variation inherent in nasalization is similar to acoustic changes associated with tongue height and jaw position, and that the effect of nasalization on F1 can be perceived as a change in tongue height.� Nasalized high vowels perceived as lower, nasalized low
vowels perceived as higher� Lingual height centralization is also well-documented
typologically for phonemic nasal vowels.� In a variety of languages, under the influence of nasalization,
high vowels are transcribed as lower and low vowels are transcribed as higher (Beddor, 1983; Hajek, 1997; Sampson, 1999).
� Delvaux (2009) shows that F2 lowering alone may help trigger the percept of nasality in French vowels.
INTERSPEECH 2013, August 29
Vowel nasalization’s problem of ambiguity
� Ambiguity in the acoustic signal� Formant changes due to nasalization or changes in
oral articulation?
� Ambiguity in perception� Formant changes due to nasalization or changes in
oral articulation?
� NB: acoustic signal alone is not enough to determine the effect of the respective articulatory contributions to the acoustic realization.
INTERSPEECH 2013, August 29
Vowel nasalization � misperception of source?
� Given the perceptual confusion between formant changes due to nasalization and formant changes due to oral articulation, there is likely a tendency for the acoustic centralization of F1 and lowering of F2 (due to nasalization) to be misperceived as oral articulatory changes.
� Following Ohala (1993, 1996), this misperception may lead to consistent, systematic changes in oral articulatory configuration as concomitants of nasal vowel phonologization.
INTERSPEECH 2013, August 29
Disambiguating NMF nasal vowels
� Previous research has observed that there are oral articulatory differences between nasal vowels and their oral counterparts in Northern Metropolitan French (NMF) � (Straka, 1965; Bricher-Labaeye, 1970; Zerling,
1984; Bothorel et al., 1986; Montagu, 2002; Delvaux et al., 2002; Carignan, 2013)
� Using AG200 and AG500, Carignan (2013) found that F1/nasalization discrepancy not predicted by lingual/labial articulation in some cases.� Pharyngealization plays a role in the acoustic
manifestation of some NMF nasal vowels?
INTERSPEECH 2013, August 29
Methodology
� Real-time MRI using partially separable functions (Liang, 2007)
� 128 × 128 voxels� Voxel resolution: 2.2 mm × 2.2 mm × 8.0 mm� ~25 fps for each of four simultaneously recorded
slices.� Four slices:
� Coronal� Oblique (velopharyngeal)� Hyperpharyngeal� Hypopharyngeal
INTERSPEECH 2013, August 29
rtMRI video (example)
� Il retape pot parfois.
� ‘He retypes pot/jar sometimes.’
� /o/� Coronal slice
INTERSPEECH 2013, August 29
Methodology (cont.)
� Three NMF female speakers� Word list containing six French lexical items, with
the target vowel in open syllable, preceded by [p]:� /pɛ/: paix ‘peace’� /pɛ̃/: pain ‘bread’� /pa/: papa ‘daddy’� /pɑ ̃/: paon ‘peacock’� /po/: pot ‘pot/jar’� /pɔ̃/: pont ‘bridge’
� Vowel boundaries segmented using noise-cancelled audio signal.
INTERSPEECH 2013, August 29
Methodology: rotation and ROI
� MR image for each vowel repetition was used to calculate an average MR image for the vowel set.
� Each MR image in a vowel set was rotated and translated with respect to the average MR image to correct for minor movements made by the speaker between repetitions.
� Average MR images for each of the six target vowel sets were used to determine the region of interest (ROI) for each MR slice for that vowel.
� The maximal bounds (i.e., x-y coordinates) of all six vowels determined the bounds of the ROI.
INTERSPEECH 2013, August 29
Methodology: rotation and ROI
� Example: Average coronal slice for NMF01.
� ROI represents the maximal coordinates of the air cavity observed for the six vowels (e.g., for the coronal slice, [ɑ̃], the lowest vowel).
� Image represents the average MR image for all six vowel sets (used for rotation and translation).
INTERSPEECH 2013, August 29
Measure 1: thArea analysis
� The cavity of interest in each MR slice was delineated by hand.
� The intensity values in the air cavity were measured in 8-bit space (values 0-255), and the maximum of the range for each slice was logged as the threshold (τ) for that slice.
� τ was used to convert each MR image into binary space: black (0) for each pixel value at or below τ, and white (1) for each pixel value above τ.
� thArea: sum of number of black pixels in ROI, multiplied by in-plane resolution (2.2 mm2)
INTERSPEECH 2013, August 29
Measure 2: pixel-articulator PCA
� ROI is determined for each vowel.� Principal Component Analysis (PCA) is used.� The intensity value of each pixel in the ROI is
included as an input the PCA model.� Reminder: high intensity values are interpreted as
flesh, low intensity values are interpreted as air.� Thus, positive loadings for PCs are interpreted as
flesh entering the frame, and negative loadings for PCs are interpreted as flesh exiting the frame.
� Result: provides a method of reliably interpreting PCs in articulatory terms (what is important?).
INTERSPEECH 2013, August 29
Measure 2: pixel-articulator PCA
� The method of using PCA, and subsequent analysis of PC1, is sufficient for explaining well-known differences between oral and nasal vowels for the velopharyngeal slice (i.e., closing of the velopharyngeal port for oral vowels).� Aside: PC2 also captures opening of the
velopharyngeal port and drag of the posterior wall.
� This presentation: � PC1 analyses for coronal, hyperpharyngeal slices� thArea analysis for hypopharyngeal slice
INTERSPEECH 2013, August 29
Coronal slice, PC1 heatmap
Interpretation all speakers: tongue raising
right
tongue
left
INTERSPEECH 2013, August 29
Coronal slice, PC1 results
� PC1 interpretation for all speakers: tongue raising
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
Carignan (2013)
INTERSPEECH 2013, August 29
Hyperpharyngeal slice, PC1 heatmap
Interpretation, all speakers: tongue retraction
posterior
anterior
right
INTERSPEECH 2013, August 29
Hyperpharyngeal slice, PC1 results
� PC1 interpretation for all speakers: tongue retraction
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
INTERSPEECH 2013, August 29
Hyperpharyngeal slice, PC1 results
� Horizontal tongue position patterns corroborate EMA data (Carignan, 2013)
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
INTERSPEECH 2013, August 29
Correlation: cor. thArea/hypo. thArea
NMF01
p<0.001, r= -0.604
•: /ɛ/
•: /ɛ̃/
•: /a/
•: /ɑ̃/
•: /o/
•: /ɔ̃/
INTERSPEECH 2013, August 29
Correlation: cor. thArea/hypo. thArea
NMF02
p<0.001, r= -0.502
•: /ɛ/
•: /ɛ̃/
•: /a/
•: /ɑ̃/
•: /o/
•: /ɔ̃/
INTERSPEECH 2013, August 29
Correlation: cor. thArea/hypo. thArea
NMF03
p<0.001, r= -0.213
•: /ɛ/
•: /ɛ̃/
•: /a/
•: /ɑ̃/
•: /o/
•: /ɔ̃/
INTERSPEECH 2013, August 29
Correlation: cor. thArea/hypo. thArea
� Negative correlation between coronal thAreaand hypopharyngealthArea.
� Corroborates that low vowels, generally, have greater pharyngeal constriction.
INTERSPEECH 2013, August 29
Pharyngeal constriction of NMF nasal vowels
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
� Hypopharngyeal thArea measure.� LME model with speaker as random effect.� All oral-nasal differences: p<0.001
INTERSPEECH 2013, August 29
Pharyngeal constriction of NMF nasal vowels
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
� Low vowel pair: less pharyngeal constriction for nasal vowel (predicted to lower F1)
� Non-low vowel pairs: greater pharyngeal constriction for nasal vowels (predicted to raise F1)
INTERSPEECH 2013, August 29
Pharyngeal constriction of NMF nasal vowels
� Remember: F1 dimension is centralized under the effect of nasalization.
� A constriction near the glottis will raise F1.� Pharyngeal constriction observed for the three
NMF oral/nasal vowel pairs is predicted to centralize the F1 dimension for the nasal vowels.
� This finding suggests that hypopharyngealaperture enhances the acoustic effect of nasalization on F1 frequency.
INTERSPEECH 2013, August 29
Lingual retraction for all nasals
� Hyperpharyngeal PC1 (i.e., tongue retraction) is greater for all nasals than their oral counterparts, for all speakers.
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
INTERSPEECH 2013, August 29
Lingual retraction for all nasals (cont.)
� Hyperpharyngeal thArea is smaller for all nasals than their oral counterparts, for all speakers.
1: /ɛ/
2: /ɛ̃/
3: /a/
4: /ɑ̃/
5: /o/
6: /ɔ̃/
INTERSPEECH 2013, August 29
Tongue retraction, F2 and nasalization in NMF
� Generally, global retraction has been observed for NMF nasal vowels compared to their oral counterparts using EMA (Carignan, 2013)
� Remember: F2 is lowered for non-back vowels under the effect of nasalization.
� “Passive” effect of nasalization� Lowered velum � “velic” constriction
� Shosted et al. (2012)
� Near an anti-node of F2� Constriction: lower F2
INTERSPEECH 2013, August 29
Tongue retraction, F2 and nasalization in NMF
� Lingual retraction observed for the three NMF nasal vowel is predicted to lower F2.
� This finding suggests that horizontal tongue position enhances the acoustic effect of nasalization on F2 frequency.
� May help explain perceptual finding from Delvaux(2009), and modeling findings from Serrurier & Badin (2008) and Feng & Castelli (1996).
INTERSPEECH 2013, August 29
Conclusion
� Our findings suggest that oral articulatory configurations enhance the effect of nasalization on F1/F2 frequencies in the production of NMF nasal vowels:� lingual articulation� hypopharyngeal constriction� labial rounding/protrusion (EMA: Carignan, 2013)
� The oral articualtion of NMF nasal vowels may be due (at least in part) to misperception of the articulatory source of changes in F1/F2 over time, rather than to mere chance.
INTERSPEECH 2013, August 29
ReferencesBeddor, P. S., “Phonological and phonetic effects of nasalization on vowel height”, Ph.D. thesis, University of Minnesota, Indiana University
Linguistics Club, 1983.Bothorel, A., Simon, P., Wioland, F., and Zerling, J.-P. Cinéradiographie des voyelles et consonnes du français. Travaux de l’Institut de
Phonétique de Strasbourg, 18, 1986.Brichler-Labaeye, C. Les voyelles françaises. Mouvements et positions articulatoires à la lumière de la radiocinématographie (coll. Bibliothèque
française et romane, série A, no. 18). Klincksieck, Paris, 1970.Carignan, C., “When nasal is more than nasal: The oral articulation of nasal vowels in two dialects of French”, Ph.D. thesis, University of Illinois
at Champaign-Urbana, 2013.Delvaux, V., “Perception du contraste de nasalité vocalique en français”, French Language Studies, 19: 25–29, 2009.Delvaux, V., Metens, T., and Soquet, A. French nasal vowels: Acoustic and articulatory properties. In Proceedings of the seventh international
conference on spoken language processing, volume 1, pages 53–56, Denver, 2002.Diehl, R. L., Kluender, K. R., Walsh, M. A., and Parker, E. M., “Auditory enhancement in speech perception and phonology”, in R. R. Hoffman
and D. S. Palermo [Eds], Cognition and the symbolic processes, Vol 3: Applied and ecological perspectives, 59–76, Erlbaum, 1991.Engwall, O., Delvaux, V., and Metens, T., “Interspeaker variation in the articulation of nasal vowels”, Online at:
http://www.speech.kth.se/prod/publications/files/1926.pdf, accessed on 11 March 2013Feng, G., and Castelli, E., “Some acoustic features of nasal and nasalized vowels: a target for vowel nasalization”, Journal of the Acoustical
Society of America, 99(6): 3694–3706, 1996.Fujimura, O., and Lindqvist, J., “Sweep-tone measurements of vocal-tract characteristics”, Journal of the Acoustical Society of America, 49: 541–
558, 1971.Hajek, J., Universals of Sound Change in Nasalization, Blackwell, 1997.Hawkins, S., and Stevens, K. N., “Acoustic and perceptual correlates of the non-nasal distinction for vowels”, Journal of the Acoustical Society of
America, 77(4):1560–1575, 1985.Kataoka, R., Warren, D., Zajaz, D. J., Mayo, R., and Lutz, R. W., “The relationship between spectral characteristics and perceived hypernasality
in children”, Journal of the Acoustical Society of America, 109(5): 2181–2189, 2001.Krakow, R. A., Beddor, P. S., and Goldstein, L. M., “Coarticulatory influences on the perceived height of nasal vowels”, Journal of the Acoustical
Society of America, 83(3): 1146–1158, 1988.Liang, Z.-P., “Spatiotemporal imaging with partially separable functions”,Biomedical Imaging: From Nano to Macro, 2007. ISBI 2007, 4th IEEE
International Symposium onBiomedical Imaging, 988–991, 2007.Montagu, J. L’articulation labiale des voyelles nasales postérieures du françcais: Comparaison entre locuteurs français et anglo-américains. In
XXIVèmes Journées d’Étude sur la Parole, pages 253–256, Nancy, 2002Ohala, J. J. “The phonetics of sound change”, in Jones, C. [Ed], Historical linguistics: Problems and perspectives, 237–78, Longman. 1993.Pruthi, T., Epsy-Wilson, C. Y., and Story, B. H., “Simulation and analysis of nasalized vowels based on magnetic resonance imaging data”,
Journal of the Acoustical Society of America, 121: 3858–3873, 2007.Sampson, R., Nasal vowel evolution in romance, Oxford University Press, 1999.Serrurier, A., and Badin., P., “A three-dimensional articulatory model of the velum and nasopharyngeal wall based on MRI and CT data”, Journal
of the Acoustical Society of America, 123(4): 2335–2355, 2008.Shosted, R., Carignan, C., and Rong, P., “Managing the distinctiveness of phonemic nasal vowels: Articulatory evidence from Hindi”, Journal of
the acoustical society of America, 131(1): 455– 465, 2012.Straka, G. Album phonétique. L’Uuniversité de Laval, Québec, 1965.Zerling, J. P. Phénomènes de nasalité et de nasalization vocaliques: Étude cinéradiographique pour deux locuteurs.Travaux de l’Institut de
Phonétique de Strasbourg, 16:241–266, 1984.
INTERSPEECH 2013, August 29